Development of an upgraded version of AZTRAN: a 3D parallel S_n transport code for light water reactor analysis

Nowadays, parallel computing in numerical simulations is more recurrent since it is becoming more and more accessible to have availability to a workstation with vast computational resources, so large problems can be modeled and executed in reasonable computational time. Usually, for this to be carried out, domain decomposition techniques are used with standard tools such as MPI and OpenMP to guarantee the portability of the code. The basic principle of domain decomposition methods is that given a domain, it is divided into several sub domains and then solved independently to obtain a much faster solution. Therefore, the field of nuclear engineering has exploited parallel computing recently in the neutronic codes that have been developed since there is great interest in performing detailed and precise simulations of nuclear reactors with acceptable computation time. Previously, due to computational limitations, the simulations were performed with many assumptions and approximations, which are still valid today. However, with present resources, one of the main objectives of nuclear engineers is to develop increasingly precise codes and efficient ones that solve the time-dependent neutron transport equation. In Mexico, the National Institute for Nuclear Research (ININ, by its acronym in Spanish) leads the AZTLAN Platform project. One of its main objectives is the development of nuclear codes to analyze nuclear reactors. These codes must follow a rigorous verification and validation process to have reliable codes and bridge the gap between similar developments around the world. One of the principal developments is AZTRAN , which is an S_N discrete-ordinates code written in Fortran 90 that solves the time-dependent multigroup neutron transport equation for cartesian geometry applying the RTN-0 nodal method. The methodology used is a source iteration scheme to approximate the angular flux of neutrons and the power method, which estimates the value of the effective multiplication factor. In this thesis, domain decomposition techniques were used to parallelize the AZTRAN code. Specifically, a spatial domain decomposition was performed with MPI and energy domain decomposition using OpenMP to take advantage of the most available computational resources. In addition, with these implementations, is possible to use a hybrid MPI-OpenMP decomposition to make parallel calculations more flexible. Finally, all the code subroutines were optimized, and some improvements were implemented, such as implementing the control rod bank movement. Now, AZTRAN can simulate 3-D transients with insertion/withdrawal of control rod banks in reasonable computation times efficiently and accurately in this updated version. In order to verify and validate the AZTRAN code, different reference problems found in the literature were used, emphasizing the well-known C5G7 Benchmark, which is proposed due to its complexity and heterogeneity. The results obtained in general are very close to the results of the references. In addition, it was found that, by increasing the angular and spatial refinement, the results are increasingly closer to those given by the MCNP reference, but the cost of using these refinements is a notable increase in computational time, so parallel computing is inherent. Concerning the parallel results, satisfactory speedups were found, especially for large problems with considerable refinement; even it found the speedup can be slightly increased for time-dependent problems. Finally, the objective of this dissertation is to provide a parallel time-dependent neutron transport code in which a transient can be simulated with whole core pin-by-pin calculation to achieve greater precision in an acceptable computational time. The goal in future developments will be to parallelize another independent variable (angular) to exploit further computational resources and coupling with a thermohydraulic code to be a full parallel high-fidelity code. (author)